High performance Computing for Palaeontology 1.3D quantitative imaging 2.Non Destructive 3.Objects up to 30 cm diameter 4.Resolutions down to 0.25  m 5.Sino-gram data size up to 10 TB 6.3D image > 10 GB 7.Challenge to do image retrieval 8.Challenge to do image segmentation

Principle of holotomography.  X-ray imaging technique with nearly coherent X-ray beams from synchrotron sources (The European Synchrotron Radiation Facility – ESRF).  Similar to a CAT scan in medicine, but using phase contrast, not absorption contrast.  Each scan consists of 1500 radiographs taken during the rotation of the sample with several different detector distance settings.  Combining these scans through a phase retrieval process allows the computing of a series of quantitative radiographic phase maps of the sample.  These are directly linked to the electron densities of the sample,  X-ray microtomography, with phase contrast can be more than a thousand times more sensitive than absorption contrast.  The application of a tomographic reconstruction algorithm on the phase maps produces a holotomographic slice stack creating a 3D virtual volume of the fossil.  The 3D processing and virtual dissections can then be made on these virtual holotomographic volumes.

Three-dimensional virtual reconstruction of the deciduous and permanent Neanderthal molars. In the image segmentation process, different physical regions in the tooth structure are digitally extracted. In this image, both enamel and dentin are rendered in transparency to show the pulp chamber and the root canals. These segmentations are amenable to considerable set of measurements and further study. Fossil teeth are usually the very best preserved treasure troves of data on the diet, feeding habits and growth chronology of their owner. They allow a rich commentary on theories of hominid behaviour and environment. Credits: Luca Bondiolli and Arnaud Mazurier. Macchiarelli et al., How Neanderthal molar teeth grew, Nature online, 22 November 2006.

Some of the insects virtually extracted from opaque amber Lak, Tafforeau, Néraudeau, Perrichot, Nel

Reconstruction of the skull of Karabo – a juvenile hominid found in a cave at Malapa in the famous Cradle of Humankind World Heritage Site close to Johannesburg. The fossil is dated at 1.9 million years old. Preliminary studies reveal fossil remnants of a shrunken brain cavity (for the very first time in hominids) also, well preserved teeth and also unerupted teeth. The 3D image requires a process called segmentation, whereby virtual objects are extracted, representing all the anatomical features of the fossil. This is currently a time consuming process, and represents a bottleneck in the analysis of the fossil data.

Proposal : High performance Computing for Palaeontology Step 1 A small group performs a scoping study 1.Multi-disciplinary – palaeontologists, chemists, geologists, computer scientists, applied mathematicians, physicists, hardware developers. 2.Review the work in this field. 3.Asses the feasibility of an Advanced Study Institute (ASI) on this project. Step 2 Should the first step be encouraging, a subgroup prepares a full proposal for an ASI 1.This involves a working visit to the STIAS facility in South Africa of 1-3 months by an interdisciplinary team. 2.There are lectures to bring the team onto the same page. 3.The team proceed to develop the project in their own field with regular team meetings. 4.The aim is to leave the STIAS facility with sufficient progress made that the project can be completed relatively soon even though future work is not always done together in the same facility. Result 1.A Set of tools representing new algorithms, which represent a massive advance in the quantitative imaging and segmentation of paleo-data. 2.The tools can be applicable more generally – medical imaging, industrial imaging, Materials Science imaging, etc.

High Performance Computing and Palaeo-Tomography The scanning of fossils at synchrotrons in X-ray radio-tomography and x-ray phase contrast tomography has revolutionised the field of palaeontology. It is possible to acquire high resolution 3-D images, where, for the latter technique the images are 3-D quantitative histograms of electron density. The resolution can be submicron, and the specimen size can be as a large as a skull. The raw sinogram can be a data file of the order of several TB. When proceeding to multiple wavelength imaging, then the data file will become tens of TB. The raw sinogram is is processed to produce the 3-D quantitative histogram. The next step, to analysis the 3-D histogram to perform the segmentation - define and virtually extract objects representing all anatomical and geological features, with due consideration to the body of knowledge in the several fields of palaeontology, anatomy, chemistry, geology, biology, applied mathematics, physics and computer science – is the bottleneck step of several years. Developing algorithms to facilitate both the sinogram processing and the segmentation processing is the goal of this project. The former is a process that requires knowledge of the physics behind the interaction of the partially coherent beam with the specimen and the image formation process on the detectors, together with the relevant mathematics and computer science to handle this scale of problem. The latter analysis is not expected to be more than semi-automatic, as it would be a sophisticated tool to be deployed under the scrutiny of a trained palaeontologist. Nonetheless, developing the tools for this analysis requires unifying the knowledge and skill from the very diverse fields mentioned above. These are both important enabling steps to properly beneficiate this new analytical capacity in palaeontology.

We propose a project to develop the mathematics for the algorithms described above – to facilitate the higher degree of automation of the analysis of paleo- tomography data with a higher degree of fidelity. The new algorithms would make full use of the quantitative nature of the data, the physics that it represents, the anatomical context, the diagenesis process and other geological, chemical and biologically relevant aspects. The outcome is to deliver a toolkit for processing based on modern computing practice. The project involves inviting a sufficient number of appropriate experts from the range of relevant disciplines, including technically competent code developers. These experts would assemble for some months in the Advanced Study Institute in Soweto (or Stellenbosch). The project is very impactful for international palaeontology. It also is significant for Southern African Heritage studies, considering that this is the cradle of humankind. The project is expected to attract significant SA funding.